Cognitive Robotics. Outline

Transcription

1 Cognitive Robotics Jeff Krichmar Lecture 4 Outline How the Body Shapes the Way We Think Ch. 4: Intelligent Systems: Properties and Principles Structured Programming Lab 4: Stay the course Follow the path. Mid-Term Project 1

2 Real World Challenges Acquisition of information takes time. Information acquired is limited. Physical devices are subject to disturbances and malfunctions. Real-world is not characterized by clearly defined, discrete states. Embodied agents must handle multiple tasks in parallel. Embodied agents must act within the dynamics of the real world. Properties of Complete Agents Subject to the laws of physics. Generate sensory stimulation through interaction with the real world. Affect the environment through behavior. Complex dynamical systems. Perform morphological computation. 2

3 Agent Design Principle 1 The Three-Constituents Principle 1) Definition of ecological niche. 2) Desired behavior. 3) Design of the agent. Design Stances Niche and behavior given. Niche given, behavior emerges. Find a niche. Scaffolding Structure our environment to make tasks easier. GPS, road signs, tools, etc. Agent Design Principle 2 The Complete-Agent When designing, we must think of the complete agent behaving in the real world. Psychology example Modular view Cognition, perception, categorization, memory, attention, social interaction, learning, emotion, etc. Complete view Different perspectives of the same set of processes. 3

4 Agent Design Principle 3: Cheap Design Agents are built to exploit Properties of the ecological niche. Characteristics of the interaction with the environment. Examples Swiss robots Passive walkers Passive Walkers Efficient Bipedal Robots Based on Passive Dynamic Walkers, Collins, et al., Science, 307: ,

5 Passive Walkers Efficient Bipedal Robots Based on Passive Dynamic Walkers, Collins, et al., Science, 307: , 2005 Agent Design Principle 4 Redundancy Intelligent agents must be designed in such a way that Their different subsystems function on the basis of different physical processes. There is partial overlap of functionality between subsystems. 5

6 Degeneracy not Redundancy Degeneracy, the ability of elements that are structurally different to perform the same function or yield the same output. Degeneracy and complexity in biological systems, GM Edelman & JA Gally, PNAS, 2001, 98(24): Degeneracy at Different Levels of Biological Organization 1. Genetic code (many different nucleotide sequences encode a polypeptide) 2. Protein fold (different polypeptides can fold to be structurally and functionally equivalent) 3. Units of transcription (degenerate initiation, termination, and splicing sites give rise to functionally equivalent mrna molecules) 4. Genes (functionally equivalent alleles, duplications, paralogs, etc., all exist) 5. Gene regulatory sequences (there are degenerate gene elements in promoters, enhancers, silencers, etc.) 6. Gene control elements (degenerate sets of transcription factors can generate similar patterns of gene expression) 7. Posttranscriptional processing (degenerate mechanisms occur in mrna processing, translocation, translation, and degradation) 8. Protein functions (overlapping binding functions and similar catalytic specificities are seen, and "moonlighting" occurs) 9. Metabolism (multiple, parallel biosynthetic and catabolic pathways exist) 10. Food sources and end products (an enormous variety of diets are nutritionally equivalent) 11. Subcellular localization (degenerate mechanisms transport cell constituents and anchor them to appropriate compartments) 12. Subcellular organelles (there is a heterogeneous population of mitochondria, ribosomes, and other organelles in every cell) 13. Cells within tissues (no individual differentiated cell is uniquely indispensable) 14. Intra- and intercellular signaling (parallel and converging pathways of various hormones, growth factors, second messengers, etc., transmit degenerate signals) 15. Pathways of organismal development (development often can occur normally in the absence of usual cells, substrates, or signaling molecules) 16. Immune responses (populations of antibodies and other antigenrecognition molecules are degenerate) 17. Connectivity in neural networks (there is enormous degeneracy in local circuitry, long-range connections, and neural dynamics) 18. Mechanisms of synaptic plasticity (changes in anatomy, presynaptic, or postsynaptic properties, etc., are all degenerate) 19. Sensory modalities (information obtained by any one modality often overlaps that obtained by others) 20. Body movements (many different patterns of muscle contraction yield equivalent outcomes) 21. Behavioral repertoires (many steps in stereotypic feeding, mating, or other social behaviors are either dispensable or substitutable) 22. Interanimal communication (there are large and sometimes nearly infinite numbers of ways to transmit the same message, a situation most obvious in language) 6

7 Agent Design Principle 5 Sensory-Motor Coordination Structured stimulation is induced through sensory-motor coordination. Agent Design Principle 6 Ecological Balance Given a task environment, there has to be a match between the complexities of the agent s sensory, motor, and neural systems. There is a balance or task distribution between morphology, materials, control, and environment. 7

8 Morphological Computation Certain processes are performed by the body that free up brain processing. Computation by Sensor Morphology The Eyebot Based on arrangement of facets in the housefly. 8

9 Agent Design Principle 7 Parallel, Loosely Coupled Processes Intelligence is emergent from a large number of parallel processes. Coordinated through embodied interaction with the environment Not Sense, think, then act. Control is parallel, asynchronous, and matched with the real-world. Agent Design Principle 8 Value Intelligent agents are equipped with a value system which constitutes a basic set of assumptions about what is good for an agent. 9

10 Value Systems Organisms adapt their behavior through value systems: Non-specific, modulatory signals to the rest of the brain. Biases the outcome of synaptic efficacy in the direction needed to satisfy global needs. Vertebrate Neuromodulatory Systems Raphe nucleus Source of serotonin Threat assessment (RED) Locus coeruleus Source of norepinephrine Novelty and saliency (ORANGE) SN & VTA Source of dopamine Reward and wanting (BLUE) Basal Forebrain Source of acetylcholine Attentional effort (GREEN) 10

11 Foraging for Different Goals Based on Internal Drives and Values Hunger Thirst Progress 11

12 Structured Programming Loops Wait for Events Switches If <condition> then [do this] Else [do that] 12